A session running inside its own worktree tries a shell command whose working directory resolves back to your main checkout, and Claude Code blocks it. Your colleague wants the quickest way past that. What is true?

The principle — an isolation guarantee that can be argued with is not one, so this check reads the destination rather than the wording and has no off switch.

Why the key is correct — three sentences settle the three halves of the question. On what is inspected: Claude Code blocks a Bash, PowerShell, or Monitor command whose working directory resolves to the main checkout. On scope, the same page adds that Claude Code blocks a Bash or Monitor command that redirects git into the main checkout — so the coverage runs in both directions, ordinary commands that land there and git redirected there. And on the way out: you can't turn this check off. Taken together, the quickest way past the block is to do the work where the session actually lives, because every route that ends in the main checkout ends the same way.

Why the others are wrong — the first reads it as a git guard when it is a working-directory guard. The second reaches for a setting that does not exist. And the third is the workaround worth naming, because it is the one a practitioner tries first: an absolute path is still resolved, and resolving is the whole check.

Remember this — the guard asks where the command would run, not how you spelled it, and the answer is not negotiable.

Sources — Anthropic's Claude Code documentation on running parallel sessions with worktrees.